We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.PR

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Probability

Title: The Chow-Robbins Game and the Catalan Triangle

Authors: John H. Elton
Abstract: The payoff in the Chow-Robbins coin-flipping game is the proportion of heads when you stop, and the goal is to maximize its expected value, by knowing when to stop. We first establish the exact optimal stop rule for positions up to a half billion flips, which suggests a fairly precise result. By using simple facts about sums involving Catalan numbers and the Catalan triangle numbers which appear in repeated backward induction, we prove that the optimal stop rule differs from the Brownian boundary minus one-half, by O(negative one-fourth power of n), and our coefficient for the negative one-fourth power for an upper bound is only slightly larger than the precise one suggested by computation. Our lower bound is of the right order in n, but with worse constant. The O(negative one-fourth power) difference was conjectured by Christensen and Fischer, with about the same coefficient, based on 100,000 flips. The proof was motivated by the idea, due to H\"aggstr\"om and W\"astlund, to use backward induction of value bounds from a horizon, which they used numerically to establish a few rigorous stop rules, and which we used with much better bounds and efficient computer use, to get rigorous stop rules for all cases up to a half billion flips. Then we use the idea algebraically instead to get a proof of the numerically-suggested asymptotics. To first get our better bounds, we use the well-known Skorohod embedding of a symmetric random walk in Brownian motion, to give a simple proof of good upper and lower bounds on the optimal stopping value in terms of the optimal value for Brownian motion. Our value upper bound was already given by Christensen and Fischer.
Subjects: Probability (math.PR)
MSC classes: 2020: 60G40 (Primary) 60G50, 60J65, 05A10, 05A19 (Secondary)
Cite as: arXiv:2205.13499 [math.PR]
  (or arXiv:2205.13499v1 [math.PR] for this version)

Submission history

From: John Elton [view email]
[v1] Thu, 26 May 2022 17:14:45 GMT (746kb)
[v2] Fri, 10 Jun 2022 17:31:23 GMT (768kb)

Link back to: arXiv, form interface, contact.